mirrors/ollama
1
0
Fork 0
mirror of https://github.com/ollama/ollama.git synced 2025-05-11 18:36:41 +02:00
Code Issues Projects Releases Packages Wiki Activity Actions

compat with upstream gguf

Browse source
This commit is contained in:
Michael Yang 2025-03-10 11:08:37 -07:00
parent c5cbe4fc2a
commit 6b32a2d549
3 changed files with 14 additions and 14 deletions
Download patch file Download diff file Expand all files Collapse all files

2
model/models/gemma3/model.go
View file

@ -62,7 +62,7 @@ func New(c ml.Config) (model.Model, error) {
TextModel: newTextModel(c),
}
slidingWindowLen := int32(c.Uint("text.attention.sliding_window"))
slidingWindowLen := int32(c.Uint("attention.sliding_window"))
m.Cache = kvcache.NewWrapperCache(kvcache.NewSWACache(slidingWindowLen, m.Shift), kvcache.NewCausalCache(m.Shift))
return &m, nil

10
model/models/gemma3/model_text.go
View file

@ -62,11 +62,11 @@ func newTextModel(c ml.Config) *TextModel {
numKVHeads: int(c.Uint("attention.head_count_kv")),
attnKeyLen: int(c.Uint("attention.key_length", 256)),
attnValLen: int(c.Uint("attention.value_length", 256)),
eps: c.Float("text.attention.layer_norm_rms_epsilon", 1e-06),
ropeLocalBase: c.Float("text.rope.local.freq_base", 10000.0),
ropeGlobalBase: c.Float("text.rope.global.freq_base", 1000000.0),
ropeScale: c.Float("text.rope.freq_scale", 1.0),
finalLogitSoftcap: c.Float("text.final_logit_softcapping", 30.0),
eps: c.Float("attention.layer_norm_rms_epsilon", 1e-06),
ropeLocalBase: c.Float("rope.local.freq_base", 10000.0),
ropeGlobalBase: c.Float("rope.global.freq_base", 1000000.0),
ropeScale: c.Float("rope.freq_scale", 1.0),
finalLogitSoftcap: c.Float("final_logit_softcapping", 30.0),
},
}